Looking At The Difference Between Econometric Modelling And Predictive Modelling

Econometric models are statistical models used in econometrics and Predictive modelling leverages statistics to predict outcomes.

That is, the only difference is who is using the statistical model. Or rather, what they are using it for. Basically, there is no difference. When individuals point out differences, it sounds a lot like a turf war. Some say econometricians don't care about prediction, they care about causality. But, why care about causality except to predict an outcome? Others say machine learning doesn't care about causality, only prediction. But that is just not true.

Predictive modelling could be seen as a larger category that encompasses econometric modelling. Typically predictive modelling involves the use of some sort of probability distribution applied as weights to various outcomes-- in discriminative prediction models; each item is placed in a predetermined category based on its weight. Alternatively-- in generative models, there are no known categories and each item is given a weight based on the likelihood that for a given output, it was the input.

Wikipedia defines statistics as, "the study of the collection, analysis, interpretation, presentation, and organization of data." Informally, I think of it as the formal study of the scientific method. It makes sense that every scientific and engineering endeavour embraces statistics. This is evident in that almost every field has a statistical branch--machine learning, biostatistics, econometrics, six sigma, signal processing, business intelligence, etc.

Econometric models can lead experts to predict the outcome of specific economic systems at different points in time. "Econometric" usually refers to a smaller set of models that utilize statistical differences and correlations to make a practical decision. Most econometric models only utilize linear operations as these are easy to understand, require little computation power, and yet give a slightly better chance of making the right decision when compared with simple intuitive decision-making.

 

 In contrast, predictive models are usually associated with some kind of Machine Learning, which cannot be done with only knowledge of statistics. These types of models require specific expertise to understand them fully and are far more accurate than simple linear regression models. They also require far more computational power.

 

And each field is accustomed to use a different subset of statistics, but we spend much too much energy sticking our noses in the air at other fields' statistics, that we fail to take advantage of the diversity of methods that could be beneficial to our problem.